More on Hyperthreading and Rendering

bgccdx wrote on 1/25/2004, 4:52 PM
For some months I have observed posts where people say that they can't get VV to use hyperthreading when rendering. As my P4 3.06Ghz PC has always shown 100% on rendering (50% on each virtual processor) I have shrugged off the issue and been thankful that my renders are very quick. Until today! I have just completed a fine cut of my visit to Central Park in NYC last week (when it was -15 deg and blanketed with 6 inches of snow!) My rough cut rendered with both processors going flat out and took around 30 mins for a 18 min vid. Today my fine cut was going to take 5hrs with only a single proc working. I was stumped, so I thought through what was the difference between the two versions that could account for the difference in render performance. I decided that it could only be the Median filter that I had applied on a single shot, so I went back, removed the median filter and re-rendered. Sure enough, it is now rendering at 100% (50% on each proc) and will now take 30 mins for my 16 min vid. I think I now understand why some people complain about VV render times. It seems if you have anything more than pretty basic filters in your project, VV falls back to a single proc.

Comments

Chienworks wrote on 1/25/2004, 5:49 PM
While all filters add to the rendering time, some are much worse than others. The Median filter is definately one of the worst.
FuTz wrote on 1/25/2004, 6:13 PM
While we're at it, making a quick list would be great, no?

Ain't GradientMap one of those? Or was it ConvolutionKernel?

A few threads now have been written about this HT stuff. I'm eager to really know if it's hype or not cause I'm planning to get a new mobo/cpu/memory bars around March or April when prices drop a bit. Seems to me that it's not as performant as promised "by the specs" (as usual; "the specs" are no more the "authority" for a lot of us now...)
rmack350 wrote on 1/25/2004, 9:11 PM
Just out of curiosity, have you compared your time with HT on and off? After all, you've still only got one processor, regardless of the virtual processors.

Rob Mack
bgccdx wrote on 1/25/2004, 11:05 PM
No, I haven't timed a render, but when both virtual processors are working, it seems to me that it is much faster than with one virtual processor doing nothing, but this is subjective rather than objective. However, my real question is, why does the addition of the median filter in one shot cause one virtual processor to do nothing? That doesn't make any sense to me. This was the point of my original post. Why does one virtual processor appear to switch off, just when it is most needed?
Randy Brown wrote on 1/26/2004, 7:00 AM
Someone please correct me if I'm wrongm but it seems I've read V4 uses one proc for video and the other for audio and of course the audio gets done way before the video (especially with filters applied). If you're checking it after the audio is finished, this theory would make sense (to me) but if you're checking it when you first start the render.... well then ignore everything I just wrote.
Randy
bgccdx wrote on 1/26/2004, 12:27 PM
I'm aware of this feature, but this isn't the situation I am referring to. In all of the instances I have spoken about, I have been speaking of rendering a video only stream. The question still remains. Why does VV appear to turn off one of the virtual processors precisely when it appears to be most needed?
roger_74 wrote on 1/26/2004, 4:20 PM
I think that's the nature of the "virtual" processors. You're still using all your processing power, nothing is lost. Real dual processors do not work the same way and you can't compare the two. In short - don't rely too much on what the CPU-usage tells you when it comes to hyperthreading.